<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Symbolic Structure on Programmer.ie: Modern AI programming</title>
    <link>http://programmer.ie/tags/symbolic-structure/</link>
    <description>Recent content in Symbolic Structure on Programmer.ie: Modern AI programming</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Sat, 05 Sep 2026 18:54:00 +0100</lastBuildDate>
    <atom:link href="http://programmer.ie/tags/symbolic-structure/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>From Vectors to Symbols — The Binding Problem Inside Neural Networks</title>
      <link>http://programmer.ie/books/models-from-first-principles/11-chapter/</link>
      <pubDate>Sat, 05 Sep 2026 18:54:00 +0100</pubDate>
      <guid>http://programmer.ie/books/models-from-first-principles/11-chapter/</guid>
      <description>&lt;h1 id=&#34;from-vectors-to-symbols--the-binding-problem-inside-neural-networks&#34;&gt;From Vectors to Symbols — The Binding Problem Inside Neural Networks&lt;/h1&gt;&#xA;&lt;p&gt;The previous chapter changed our scale of analysis.&lt;/p&gt;&#xA;&lt;p&gt;We stopped asking only what architecture a model had and started asking where the computation lived.&lt;/p&gt;&#xA;&lt;p&gt;Was capability coming from new parameters?&lt;/p&gt;&#xA;&lt;p&gt;More recurrent steps?&lt;/p&gt;&#xA;&lt;p&gt;More sampled trajectories?&lt;/p&gt;&#xA;&lt;p&gt;A selector?&lt;/p&gt;&#xA;&lt;p&gt;A verifier?&lt;/p&gt;&#xA;&lt;p&gt;Different post-training?&lt;/p&gt;&#xA;&lt;p&gt;That wider view exposed a second question.&lt;/p&gt;&#xA;&lt;p&gt;Suppose the computation works.&lt;/p&gt;&#xA;&lt;p&gt;Suppose a neural network solves arithmetic, manipulates code, follows grammatical structure, or reasons over logical relations.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
